Multicarrier modulation (or Discrete Multitone Modulation (DMT)) is a transmission method that is being widely used for communication over difficult media. Multicarrier modulation divides the transmission frequency band into multiple subchannels (carriers), with each carrier individually modulating a bit or a collection of bits. A transmitter modulates an input data stream containing information bits with one or more carriers and transmits the modulated information. A receiver demodulates all the carriers in order to recover the transmitted information bits as an output data stream.
Multicarrier modulation has many advantages over single carrier modulation. These advantages include, for example, a higher immunity to impulse noise, a lower complexity equalization requirement in the presence of multipath, a higher immunity to narrow band interference, a higher data rate and bandwidth flexibility. Multicarrier modulation is being used in many applications to obtain these advantages, as well as for other reasons. Applications include Asymmetric Digital Subscriber Line (ADSL) systems, Wireless LAN systems, Power Line communications systems, and other applications. ITU standards G.992.1 and G.992.2 and the ANSI T1.413 standard specify standard implementations for ADSL transceivers that use multicarrier modulation.
The block diagram 100 for a standard compliant ADSL DMT transmitter known in the art is shown in FIG. 1. FIG. 1 shows three layers: the Modulation layer 110, the Framer/FEC layer 120, and the ATM TC layer 140, which are described below.
The Modulation layer 110 provides functionality associated with DMT modulation. DMT modulation is implemented using an Inverse Discrete Fourier Transform (IDFT) 112. The IDFT 112 modulates bits from the Quadrature Amplitude Modulation (QAM) 114 encoder into the multicarrier subchannels. ADSL multicarrier transceivers modulate a number of bits on each subchannel, the number of bits depending on the Signal to Noise Ratio (SNR) of that subchannel and the Bit Error Rate (BER) requirement of the link. For example, if the required BER is 1×10−7 (i.e., one bit in ten million is received in error on average) and the SNR of a particular subchannel is 21.5 dB, then that subchannel can modulate 4 bits, since 21.5 dB is the required SNR to transmit 4 QAM bits with a 1×10−7 BER. Other subchannels can have a different SNR and therefore may have a different number of bits allocated to them at the same BER. The ITU and ANSI ADSL standards allow up to 15 bits to be modulated on one carrier.
A table that specifies how many bits are allocated to each subchannel for modulation in one DMT symbol is called a Bit Allocation Table (BAT). A DMT symbol is the collection of analog samples generated at the output of the IDFT by modulating the carriers with bits according to the BAT. The BAT is the main parameter used in the Modulation layer 110 of FIG. 1. The BAT is used by the QAM 114 and IDFT 112 blocks for encoding and modulation. Table 1 shows an example of a BAT for a DMT system with 16 subchannels.
TABLE 1Example of BAT for multicarrier system with 16 subchannelsSubchannelBits perNumberSubchannel152933425460758798103110125136148154163Total bits80Per DMTsymbol
In ADSL systems the DMT symbol rate is approximately 4 kHz. This means that a new DMT symbol modulating a new set of bits, using the modulation BAT, is transmitted every 250 microseconds. If the BAT in table 1, which specifies 80 bits modulated in one DMT symbol, were used at a 4 kHz DMT symbol rate the bit rate of the system would be 4000*80=320 kilobits per second (kbps). The BAT determines the data rate of the system and is dependent on the transmission channel characteristics, i.e. the SNR of each subchannel in the multicarrier system. A channel with low noise (high SNR on each subchannel) will have many bits modulated on each DMT carrier and will thus have a high bit rate. If the channel conditions are poor, the SNR will be low and the bits modulated on each carrier will be few, resulting in a low system bit rate. As can be seen in Table 1, some subchannels may actually modulate zero bits. An example is the case when a narrow band interferer (such as AM broadcast radio) is present at a subchannel's frequency and causes the SNR in that subchannel to be too low to carry any information bits.
The ATM TC layer 140 includes an Asynchronous Transfer Mode Transmission Convergence (ATM TC) block 142 that transforms bits and bytes in cells into frames.
The next layer in an ADSL system is the Frame/FEC layer 120, which provides functionality associated with preparing a stream of bits for modulation, as shown in FIG. 1. This layer contains the Interleaving (INT) block 122, the Forward Error Correction (FEC) block 124, the scrambler (SCR) block 126, the Cyclic Redundancy Check (CRC) block 128 and the ADSL Framer block 130. Interleaving and FEC coding provide impulse noise immunity and a coding gain. The FEC 124 in the standard ADSL system is a Reed-Solomon (R-S) code. The scrambler 126 is used to randomize the data bits. The CRC 128 is used to provide error detection at the receiver. The ADSL Framer 130 frames the received bits from the ATM framer 142. The ADSL framer 130 also inserts and extracts overhead bits from module 132 for modem to modem overhead communication channels (known as EOC and AOC channels in the ADSL standards).
The key parameters in the Framer/FEC layer 120 are the size of the R-S codeword, the size (depth) of the interleaver (measured in number of R-S codewords) and the size of the ADSL frame. As examples, a typical size for an R-S codeword may be 216 bytes, a typical size for interleaver depth may be 64 codewords, and the typical size of the ADSL frame may be 200 bytes. It is also possible to have an interleaving depth equal to one, which is equivalent to no interleaving. In order to recover the digital signal that was originally prepared for transmission using a transmitter as discussed above, it is necessary to deinterleave the codewords by using a deinterleaver that performs the inverse process to that of the interleaver, with the same depth parameter. In the current ADSL standards there is a specific relationship between all of these parameters in a DMT system. Specifically, the BAT size, NBAT (total number of bits in a DMT symbol) is fixed to be an integer divisor of the R-S codeword size, NFEC, as expressed in equation (1):NFEC=S×NBAT, where S is a positive integer greater than 0.  (1)
This constraint can also be expressed as: One R-S codeword contains an integer number of DMT symbols. The R-S codeword contains data bytes and parity (checkbytes). The checkbytes are overhead bytes that are added by the R-S encoder and are used by the R-S decoder to detect and correct bit errors. There are R checkbytes in a R-S codeword. Typically, the number of checkbytes is a small percentage of the overall codeword size, e.g., 8%. Most channel coding methods are characterized by their coding gain, which is defined as the system performance improvement (in dB) provided by the code when compared to an uncoded system. The coding gain, of the R-S codeword depends on the number of checkbytes and the R-S codeword size. A large R-S codeword (greater than 200 bytes in a DMT ADSL system) along with a 16 checkbytes (8% of 200 bytes) will provide close to the maximum coding gain of 4 dB. If the codeword size is smaller and/or the percentage of checkbyte overhead is high (e.g. >30%) the coding gain may be very small or even negative. In general, it is best to have the ADSL system operating with the largest possible R-S codeword (the maximum possible is 255 bytes) and approximately 8% redundancy.
There is also a specific relationship between the number of bytes in an ADSL frame, NFRAME, and the R-S codeword size, NFEC that is expressed in equation (2):NFEC=S×NFRAME+R; where R is the number of R-S checkbytes in a codeword and S is the same positive integer in Equation (1).  (2)It is apparent from equating the right-hand sides of equations (1) and (2) that the relationship expressed in equation (3) resultsNBAT=NFRAME+R/S.  (3)The ADSL standard requires that the ratio (R/S) is an integer, i.e. there is an integer number of R-S checkbytes in every DMT-symbol (NBAT). As described above, ADSL frames contain overhead bytes (not part of the payload) that are used for modem to modem communications. A byte in an ADSL frame that is used for the overhead channel cannot be used for the actual user data communication, and therefore the user data rate decreases accordingly. The information content and format of these channels is described in the ITU and ANSI standards. There are several framing modes defined in ADSL standards. Depending on the framing mode, there are more or fewer overhead bytes in one ADSL frame. For example, standard Framing Mode 3 has 1 overhead byte per ADSL frame.
Equations (1), (2) and (3) demonstrate that the parameter restrictions imposed by the standards result in the following conditions:    1. All DMT symbols have a fixed number of overhead framing bytes that are added at the ADSL framer. For example, in framing mode #3 there is 1 overhead framing byte per DMT symbol.    2. There is a minimum of 1 R-S checkbyte per DMT symbol.    3. The maximum number of checkbytes according to ITU Standard G.992.2 (8) and ITU Standards G.992.2 and T1.413 (16) limits the maximum codeword size to 8*NBAT for G.992.2, and to 16*NBAT for G.992.1 and T1.413.    4. An ADSL modem cannot change the number of bits in a DMT symbol (NBAT) without making the appropriate changes to the number of bytes in a R-S codeword (NFEC) and an ADSL frame (NFRAME).The above four restrictions cause performance limitations in current ADSL systems.
In particular, because of condition #1 every DMT symbol has a fixed number of overhead framing bytes. This is a problem when the data rate is low and the overhead framing bytes consume a large percentage of the possible throughput resulting in a lower payload. For example, if the date rate supported by the line is 6.144 Mbps, this will result in a DMT symbol with about 192 bytes per symbol (192*8*4000=6144000 bps). In this case, one overhead framing byte would consume 1/192 or about 0.5% of the available throughout. But if the date rate is 128 kbps or 4 bytes per symbol the overhead framing byte will consume ¼ or 25% of the available throughput. Clearly this is undesirable.
Condition #2 will cause the same problems as condition # 1. In this case, the overhead framing byte is replaced by the R-S checkbyte.
Condition #3 will not allow the construction of large codewords when the data rate is low. R-S codewords in ADSL can have a maximum of 255 bytes. The maximum coding gain is achieved when the codeword size is near the maximum 255 bytes. When the data rate is low, e.g. 128 kbps or 4 bytes per symbol, the maximum codeword size will be 8*4=32 bytes for G.992.2 systems and 16*4=64 bytes for G.992.1 and T1.413 systems. In this case the coding gain will be substantially lower than for large codewords approaching 255 bytes.
In general, if the data rate is low, e.g. 128 kbps or 4 byte per symbol, the above conditions will result in 1 byte being used for overhead framing, and 1 byte being consumed by a R-S checkbyte. Therefore 50% of the available throughput will not be used for payload and the R-S codeword size will be at most 64 bytes, resulting in negligible coding gain.
Condition #4 effects the ability of the modem to adapt its transmission parameters on-line in a dynamic manner.
G.992.1 and T1.413 specify a mechanism to do on-line rate adaptation, called Dynamic Rate Adaptation (DRA), but it is clearly stated in these standards that the change in data rate will not be seamless. In general current ADSL DMT modems use Bit Swapping and dynamic rate adaptation (DRA) as methods for on-line adaptation to channel changes. Bit Swapping is specified in the JTU and ANSI standards as method for modifying the number of bits allocated to a particular carrier. Bit Swapping is seamless, i.e., it does not result in an interruption in data transmission and reception. But, Bit Swapping does not allow the changing of data rates. Bit Swapping only allows the changing the number of bits allocated to carriers while maintaining the same data rate. This is equivalent to changing the entries in the BAT table without allowing the total number of bits (N.sub.BAT) in the BAT to increase or decrease.
DRA enables a change in data rate, but is not seamless. DRA is also very slow because it requires the modem located in the Central Office (CO) to make the final decision on the data rate configuration. This model, with (the CO being the master), is common among ADSL modems that are designed to provide a service offered by the telephone company, and controlled by the telephone company.
Both Bit Swapping and DRA use a specific protocol that is specified in ANSI T1.413, G.992.1 and G.992.2 for negotiating the change. This protocol negotiates the parameters using messages that are sent via an AOC channel, which is an embedded channel. This protocol is sensitive to impulse noise and high noise levels. If the messages are corrupted the transmitter and receiver can enter a state where they are using different transmission parameters (e.g., BAT, data rate, R-S codeword length, interleaver depth, etc). When two communication modems enter a state of mismatched transmission parameters, data will be received in error and the modems will eventually be required to take drastic measures (such as full reinitialization) in order to restore error free transmission. Drastic measures such as full reinitialization will result in the service being dropped for approximately 10 seconds, which is the time required for a standards compliant ADSL modem to complete a full initialization.
A transceiver has both a transmitter and a receiver. The receiver includes the receiver equivalent blocks of the transmitter shown in FIG. 1. The receiver has modules that include a decoder, a deinterleaver and a demodulator. In operation, the receiver accepts a signal in analog form that was transmitted by a transmitter, optionally amplifies the signal in an amplifier, filters the signal to remove noise components and to separate the signal from other frequencies, converts the analog signal to a digital signal through the use of an analog to digital converter, demodulates the signal to generate the received bits stream from the carrier subchannels by the use of a demodulator, deinterleaves the bits stream by the use of a deinterleaver, performs the FEC decoding to correct errors in the bit stream by use of an FEC decoder, descrambles the bit stream by use of a descrambler, and detects bit errors in the bit stream by use of a CRC. Various semiconductor chip manufacturers supply hardware and software that can perform the functions of a transmitter or a receiver, or both.
It is therefore apparent that there is a need for an improved DMT transmission system. It is therefore a principle object of the invention to provide an improved DMT transmission system that overcomes the problems discussed above.